Energy consumption prediction method and system for multifunctional intelligent building

Abstract

The invention discloses an energy consumption prediction method and system for a multifunctional intelligent building, and relates to the technical field of building energy consumption management, and the method comprises the steps: S1, setting a plurality of sensors to collect the multi-source heterogeneous data of the current multifunctional intelligent building; s2, constructing the collected multi-source heterogeneous data into a data association graph, mining a deep relationship between the data by using a graph neural network, and obtaining an enhanced energy consumption related feature set; s3, combing causal relationships in the feature set by using a causal inference model, calculating feature weights and outputting energy consumption predicted values; self-adaptive optimization of the model is realized by setting a reward function; s4, constructing a plurality of energy consumption scene simulation schemes, simulating and predicting energy consumption for multiple times in each scene through a simulation method, and determining probability distribution obeyed by an energy consumption prediction result; and S5, establishing a real-time energy consumption data feedback loop, and comparing and analyzing actual energy consumption data and a prediction result.

Description

Technical Field

[0001] The present invention belongs to the technical field of building energy consumption management, and in particular relates to an energy consumption prediction method and system for a multifunctional intelligent building. Background Art

[0002] In today's society, multifunctional intelligent buildings are becoming increasingly popular. Their internal functions are complex and diverse, covering multiple areas such as office, commercial, residential, and various public service facilities. This makes the building energy consumption situation affected by many complex and dynamically changing factors. Traditional energy consumption prediction methods often focus on the use of a single type of data or simple model construction, which makes it difficult to accurately capture the complex changes in the energy consumption of multifunctional intelligent buildings, resulting in a large deviation between the prediction results and the actual energy consumption, and unable to provide a reliable basis for effective energy management and energy-saving measures. Therefore, we propose an energy consumption prediction method and system for multifunctional intelligent buildings. Summary of the invention

[0003] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0004] The present invention is an energy consumption prediction method and system for a multifunctional intelligent building, comprising step S1: setting a plurality of sensors to collect multi-source heterogeneous data of the current multifunctional intelligent building;

[0005] Step S2: construct the collected multi-source heterogeneous data into a data association graph, use the graph neural network to mine the deep relationship between the data, obtain the fused feature vector, input the fused feature vector into the generative adversarial network, and obtain the enhanced energy consumption related feature set;

[0006] Step S3: Use the causal inference model to sort out the causal relationship in the feature set and embed it into the improved deep learning time series model. The model is based on the gated recurrent unit to calculate the feature weights and output energy consumption prediction values; and realize model adaptive optimization by setting the reward function;

[0007] Step S4: construct multiple energy consumption scenario simulation schemes, simulate and predict energy consumption multiple times under each scenario through simulation methods, determine the probability distribution obeyed by the energy consumption prediction results, and finally output the energy consumption prediction results in the form of probability intervals;

[0008] Step S5: Establish a real-time energy consumption data feedback loop to compare and analyze the actual energy consumption data with the predicted results. When the deviation exceeds the preset threshold, the model re-evaluation and parameter adjustment mechanism is triggered.

[0009] Furthermore, in step S1: by installing a variety of sensors in the multifunctional building, the changes in temperature and humidity in the building, the number and behavior of people, the working conditions of large electromechanical equipment and the data of surrounding meteorological monitoring stations are collected.

[0010] Furthermore, the step S2 includes the following sub-steps:

[0011] Step S21: Assume that the collected multi-source heterogeneous data is represented as X=[x1,x2,…xn], where xi represents the data point collected by the i-th sensor, the total number of data points is n, and construct a data association graph G=(V,E), where the node set V corresponds to different data points xi, and the edge set E represents the relationship between the data points;

[0012] The information propagation update formula of the graph neural network is as follows:

[0013] For node vi∈v, at the tth iteration, its feature representation is The update formula is:

[0014]

[0015] Among them, N(i) represents the set of neighbor nodes of node vi, cij represents the normalization constant, Wt represents the weight matrix of the tth iteration, bt represents the bias term, and σ is the activation function;

[0016] After T iterations, the fused node feature vector is obtained. This is used as the initial fusion feature;

[0017] Step S22: The above fused feature vector H is input into the generative adversarial network, and the generator G generates a new feature Hfake that is close to the real energy consumption related feature distribution. The generation process can be expressed as:

[0018] Hfake = G(z;θG);

[0019] Where z represents the random noise vector and θG represents the parameters of the generator;

[0020] The discriminator D is responsible for distinguishing the true fusion features from H and the fake features Hfake generated by the generator. The probability values ​​of its discriminant outputs D(H) and D(Hfake) are calculated by the following formula:

[0021]

[0022] Where fD represents the decision function of the discriminator, and θD represents the parameters of the discriminator;

[0023] The generator G and the discriminator G are trained with the following adversarial loss function LG:

[0024]

[0025] Through multiple iterative training, the feature Hfake generated by the generator is continuously approached to the real feature H, and finally a representative and discriminative energy consumption related feature set He is obtained.

[0026] Furthermore, the step S3 includes the following sub-steps:

[0027] Step S31: Use the causal inference model to sort out the features with causal relationships in the He feature set. Assume that the causal relationship between variable X and energy consumption variable Y is represented by conditional probability P(YX), and its calculation formula is:

[0028]

[0029] Where P(YX) represents the probability of factor X appearing under the condition of energy consumption Y, P(Y) is the prior probability of energy consumption variable Y, and P(X) is the probability of factor X;

[0030] By analyzing these causal relationships, a preliminary energy consumption causal relationship network structure is constructed and embedded into an improved deep learning time series model;

[0031] Step S32: The model is based on a gated recurrent unit. For the gated recurrent unit, the calculation formulas for the update gate zt, reset gate rt and hidden state ht are as follows:

[0032] zt=σ(Wz·[xt,ht-1]+bz);

[0033] rt = σ(Wr · [xt, ht-1] + br);

[0034] ht=(1-zt)·ht-1+zt·tanh(W·[rt·ht-1,xt]+b);

[0035] Where xt is the input feature vector at time t, Wz, Wr, W are the corresponding weight matrices, bz, br, b are bias terms, and σ is the sigmoid function;

[0036] The attention mechanism is used to calculate the weight ati of the feature at each moment for energy consumption prediction. The calculation formula is as follows:

[0037]

[0038] Where v, Wa, ba represent the parameters of the attention mechanism, n represents the dimension of the input feature vector, and eti represents the intermediate calculation result;

[0039] The final energy consumption prediction output yt is obtained by linearly transforming the weighted hidden state:

[0040]

[0041] Where Wy and by represent the weight matrix and bias term of the output layer;

[0042] At the same time, the structure and parameters of the model can be adaptively adjusted according to the new input data and prediction error feedback. Through an evolutionary learning algorithm based on reinforcement learning, the reward function R is set to represent the matching degree between the prediction result and the actual energy consumption. According to the strategy π, action a is taken to optimize the model by maximizing the expected reward J(π). The calculation formula is:

[0043]

[0044] Where γ is a discount factor that weighs the importance of future rewards against current rewards, and Rt is the reward value at time t.

[0045] Furthermore, in step S4: based on historical data and real-time collected data, combined with urban development planning, seasonal change trends and possible special events, multiple energy consumption scenario simulation schemes are constructed, and multiple simulation predictions are performed on the model under each scenario;

[0046] Assume that the energy consumption prediction result y obeys a certain probability distribution, and obtains m prediction values ​​y1, y2, …, ym, through multiple simulations. The calculation formulas of the mean μ and variance σ2 are as follows:

[0047]

[0048] The output energy consumption prediction results are presented in the form of probability intervals.

[0049] Furthermore, in step S5: by establishing a real-time energy consumption data feedback loop, the actual energy consumption data is compared and analyzed with the predicted results. Once it is found that the deviation exceeds the preset threshold, the model re-evaluation and parameter adjustment mechanism is immediately triggered.

[0050] The present invention has the following beneficial effects:

[0051] 1. The present invention uses clear data preprocessing formulas and feature extraction and fusion formulas of the attention mechanism to carefully and accurately mine the spatiotemporal characteristics of the energy consumption data of multifunctional intelligent buildings, overcome the limitations of traditional methods, provide strong support for accurate prediction, and more comprehensively capture the complex correlations between various factors affecting energy consumption in the time and space dimensions, so that subsequent prediction models can estimate energy consumption based on more representative feature information.

[0052] 2. The present invention significantly improves the model's ability to fit complex nonlinear energy consumption relationships, reduces prediction errors, and enhances the generalization ability of the model by systematically adjusting key parameters of membership functions and rule output functions, so that it can show good and stable prediction performance when predicting the energy consumption of smart buildings under different functional layouts, different operating conditions, and changing environments. It accurately updates model parameters based on newly generated real-time data, and effectively responds to changes in energy consumption caused by factors such as changes in the internal and external environment of the building, equipment updates, and changes in personnel activity patterns. The prediction model always closely tracks the actual energy consumption of the building, maintains a high prediction accuracy, and ensures that reliable prediction results are continuously provided for building energy management.

[0053] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0055] Figure 1 The present invention is a flowchart of an energy consumption prediction method and system for a multifunctional intelligent building. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] See also Figure 1 As shown, the present invention is a method and system for predicting energy consumption of a multifunctional intelligent building, comprising step S1: setting a plurality of sensors to collect multi-source heterogeneous data of the current multifunctional intelligent building;

[0058] Step S2: construct the collected multi-source heterogeneous data into a data association graph, use the graph neural network to mine the deep relationship between the data, obtain the fused feature vector, input the fused feature vector into the generative adversarial network, and obtain the enhanced energy consumption related feature set;

[0059] Step S3: Use the causal inference model to sort out the causal relationship in the feature set and embed it into the improved deep learning time series model. The model is based on the gated recurrent unit to calculate the feature weights and output energy consumption prediction values; and realize model adaptive optimization by setting the reward function;

[0060] Step S4: construct multiple energy consumption scenario simulation schemes, simulate and predict energy consumption multiple times under each scenario through simulation methods, determine the probability distribution obeyed by the energy consumption prediction results, and finally output the energy consumption prediction results in the form of probability intervals;

[0061] Step S5: Establish a real-time energy consumption data feedback loop to compare and analyze the actual energy consumption data with the predicted results. When the deviation exceeds the preset threshold, the model re-evaluation and parameter adjustment mechanism is triggered.

[0062] In step S1: by installing a variety of sensors in the multifunctional building, the changes in temperature and humidity in the building, the number and behavior of people, the working conditions of large electromechanical equipment and the data of surrounding meteorological monitoring stations are collected.

[0063] Step S2 includes the following sub-steps:

[0064] Step S21: Assume that the collected multi-source heterogeneous data is represented as X=[x1,x2,…xn], where xi represents the data point collected by the i-th sensor, the total number of data points is n, and construct a data association graph G=(V,E), where the node set V corresponds to different data points xi, and the edge set E represents the relationship between the data points;

[0065] The information propagation update formula of the graph neural network is as follows:

[0066] For node vi∈v, at the tth iteration, its feature representation is The update formula is:

[0067]

[0068] Among them, N(i) represents the set of neighbor nodes of node vi, cij represents the normalization constant, Wt represents the weight matrix of the tth iteration, bt represents the bias term, and σ is the activation function;

[0069] After T iterations, the fused node feature vector is obtained. This is used as the initial fusion feature;

[0070] Step S22: The above fused feature vector H is input into the generative adversarial network, and the generator G generates a new feature Hfake that is close to the real energy consumption related feature distribution. The generation process can be expressed as:

[0071] Hfake = G(z;θG);

[0072] Where z represents the random noise vector and θG represents the parameters of the generator;

[0073] The discriminator D is responsible for distinguishing the true fusion features from H and the fake features Hfake generated by the generator. The probability values ​​of its discriminant outputs D(H) and D(Hfake) are calculated by the following formula:

[0074]

[0075] Where fD represents the decision function of the discriminator, and θD represents the parameters of the discriminator;

[0076] The generator G and the discriminator G are trained with the following adversarial loss function LG:

[0077]

[0078] Through multiple iterative training, the feature Hfake generated by the generator is continuously approached to the real feature H, and finally a representative and discriminative energy consumption related feature set He is obtained.

[0079] Step S3 includes the following sub-steps:

[0080] Step S31: Use the causal inference model to sort out the features with causal relationships in the He feature set. Assume that the causal relationship between variable X and energy consumption variable Y is represented by conditional probability P(YX), and its calculation formula is:

[0081]

[0082] Where P(YX) represents the probability of factor X appearing under the condition of energy consumption Y, P(Y) is the prior probability of energy consumption variable Y, and P(X) is the probability of factor X;

[0083] By analyzing these causal relationships, a preliminary energy consumption causal relationship network structure is constructed and embedded into an improved deep learning time series model;

[0084] Step S32: The model is based on a gated recurrent unit. For the gated recurrent unit, the calculation formulas for the update gate zt, reset gate rt and hidden state ht are as follows:

[0085] zt=σ(Wz·[xt,ht-1]+bz);

[0086] rt = σ(Wr · [xt, ht-1] + br);

[0087] ht=(1-zt)·ht-1+zt·tanh(W·[rt·ht-1,xt]+b);

[0088] Where xt is the input feature vector at time t, Wz, Wr, W are the corresponding weight matrices, bz, br, b are bias terms, and σ is the sigmoid function;

[0089] The attention mechanism is used to calculate the weight ati of the feature at each moment for energy consumption prediction. The calculation formula is as follows:

[0090]

[0091] Where v, Wa, ba represent the parameters of the attention mechanism, n represents the dimension of the input feature vector, and eti represents the intermediate calculation result;

[0092] The final energy consumption prediction output yt is obtained by linearly transforming the weighted hidden state:

[0093]

[0094] Where Wy and by represent the weight matrix and bias term of the output layer;

[0095] At the same time, the structure and parameters of the model can be adaptively adjusted according to the new input data and prediction error feedback. Through an evolutionary learning algorithm based on reinforcement learning, the reward function R is set to represent the matching degree between the prediction result and the actual energy consumption. According to the strategy π, action a is taken to optimize the model by maximizing the expected reward J(π). The calculation formula is:

[0096]

[0097] Where γ is a discount factor that weighs the importance of future rewards against current rewards, and Rt is the reward value at time t.

[0098] In step S4: based on historical data and real-time collected data, combined with urban development planning, seasonal change trends and possible special events, multiple energy consumption scenario simulation schemes are constructed, and multiple simulation predictions are performed on the model under each scenario;

[0099] Assume that the energy consumption prediction result y obeys a certain probability distribution, and obtains m prediction values ​​y1, y2, …, ym, through multiple simulations. The calculation formulas of the mean μ and variance σ2 are as follows:

[0100]

[0101] The output energy consumption prediction results are presented in the form of probability intervals.

[0102] In step S5: by establishing a real-time energy consumption data feedback loop, the actual energy consumption data is compared and analyzed with the predicted results. Once the deviation is found to exceed the preset threshold, the model re-evaluation and parameter adjustment mechanism is immediately triggered.

[0103] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0104] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method and system for predicting energy consumption of a multifunctional intelligent building, characterized in that: The steps include: Step S1: Setting up a variety of sensors to collect multi-source heterogeneous data of current multifunctional intelligent buildings; Step S2: construct the collected multi-source heterogeneous data into a data association graph, use the graph neural network to mine the deep relationship between the data, obtain the fused feature vector, input the fused feature vector into the generative adversarial network, and obtain the enhanced energy consumption related feature set; Step S3: Use the causal inference model to sort out the causal relationship in the feature set and embed it into the improved deep learning time series model. The model is based on the gated recurrent unit to calculate the feature weights and output energy consumption prediction values; and realize model adaptive optimization by setting the reward function; Step S4: construct multiple energy consumption scenario simulation schemes, simulate and predict energy consumption multiple times under each scenario through simulation methods, determine the probability distribution obeyed by the energy consumption prediction results, and finally output the energy consumption prediction results in the form of probability intervals; Step S5: Establish a real-time energy consumption data feedback loop to compare and analyze the actual energy consumption data with the predicted results. When the deviation exceeds the preset threshold, the model re-evaluation and parameter adjustment mechanism is triggered.

2. The energy consumption prediction method and system for a multifunctional intelligent building according to claim 1, characterized in that: In step S1: by installing a variety of sensors in the multifunctional building, the changes in temperature and humidity in the building, the number and behavior of people, the working conditions of large electromechanical equipment and the data of surrounding meteorological monitoring stations are collected.

3. The energy consumption prediction method and system for a multifunctional intelligent building according to claim 1, characterized in that: The step S2 includes the following sub-steps: Step S21: Assume that the collected multi-source heterogeneous data is represented as X=[x1,x2,…xn], where xi represents the data point collected by the i-th sensor, the total number of data points is n, and construct a data association graph G=(V,E), where the node set V corresponds to different data points xi, and the edge set E represents the relationship between the data points; The information propagation update formula of the graph neural network is as follows: For node vi∈v, at the tth iteration, its feature representation is The update formula is: Among them, N(i) represents the set of neighbor nodes of node vi, cij represents the normalization constant, Wt represents the weight matrix of the tth iteration, bt represents the bias term, and σ is the activation function; After T iterations, the fused node feature vector is obtained. This is used as the initial fusion feature; Step S22: The above fused feature vector H is input into the generative adversarial network, and the generator G generates a new feature Hfake that is close to the real energy consumption related feature distribution. The generation process can be expressed as: Hfake = G(z;θG); Where z represents the random noise vector and θG represents the parameters of the generator; The discriminator D is responsible for distinguishing the true fusion features from H and the fake features Hfake generated by the generator. The probability values ​​of its discriminant outputs D(H) and D(Hfake) are calculated by the following formula: Where fD represents the decision function of the discriminator, and θD represents the parameters of the discriminator; The generator G and the discriminator G are trained with the following adversarial loss function LG: Through multiple iterative training, the feature Hfake generated by the generator is continuously approached to the real feature H, and finally a representative and discriminative energy consumption related feature set He is obtained.

4. The energy consumption prediction method and system for a multifunctional intelligent building according to claim 1, characterized in that: The step S3 includes the following sub-steps: Step S31: Use the causal inference model to sort out the features with causal relationships in the He feature set. Assume that the causal relationship between variable X and energy consumption variable Y is represented by conditional probability P(YX), and its calculation formula is: Where P(YX) represents the probability of factor X appearing under the condition of energy consumption Y, P(Y) is the prior probability of energy consumption variable Y, and P(X) is the probability of factor X; By analyzing these causal relationships, a preliminary energy consumption causal relationship network structure is constructed and embedded into an improved deep learning time series model; Step S32: The model is based on a gated recurrent unit. For the gated recurrent unit, the calculation formulas for the update gate zt, reset gate rt and hidden state ht are as follows: zt=σ(Wz·[xt,ht-1]+bz); rt = σ(Wr · [xt, ht-1] + br); ht=(1-zt)·ht-1+zt·tanh(W·[rt·ht-1,xt]+b); Where xt is the input feature vector at time t, Wz, Wr, W are the corresponding weight matrices, bz, br, b are bias terms, and σ is the sigmoid function; The attention mechanism is used to calculate the weight ati of the feature at each moment for energy consumption prediction. The calculation formula is as follows: Where v, Wa, ba represent the parameters of the attention mechanism, n represents the dimension of the input feature vector, and eti represents the intermediate calculation result; The final energy consumption prediction output yt is obtained by linearly transforming the weighted hidden state: Where Wy and by represent the weight matrix and bias term of the output layer; At the same time, the structure and parameters of the model can be adaptively adjusted according to the new input data and prediction error feedback. Through an evolutionary learning algorithm based on reinforcement learning, the reward function R is set to represent the matching degree between the prediction result and the actual energy consumption. According to the strategy π, action a is taken to optimize the model by maximizing the expected reward J(π). The calculation formula is: Where γ is a discount factor that weighs the importance of future rewards against current rewards, and Rt is the reward value at time t.

5. The energy consumption prediction method and system for a multifunctional intelligent building according to claim 1, characterized in that: In the step S4: based on historical data and real-time collected data, combined with urban development planning, seasonal change trends and possible special events, multiple energy consumption scenario simulation schemes are constructed, and multiple simulation predictions are performed on the model under each scenario; Assume that the energy consumption prediction result y obeys a certain probability distribution, and obtains m prediction values ​​y1, y2, …, ym, through multiple simulations. The calculation formulas of the mean μ and variance σ2 are as follows: The output energy consumption prediction results are presented in the form of probability intervals.

6. The energy consumption prediction method and system for a multifunctional intelligent building according to claim 1, characterized in that: In step S5: by establishing a real-time energy consumption data feedback loop, the actual energy consumption data is compared and analyzed with the predicted results. Once it is found that the deviation exceeds the preset threshold, the model re-evaluation and parameter adjustment mechanism is immediately triggered.

Images